Information theoretical methods for complex network structure reconstruction
© Hernández-Lemus and Siqueiros-García; licensee Springer. 2013
Received: 12 October 2012
Accepted: 19 February 2013
Published: 8 April 2013
Skip to main content
© Hernández-Lemus and Siqueiros-García; licensee Springer. 2013
Received: 12 October 2012
Accepted: 19 February 2013
Published: 8 April 2013
Complex networks seem to be ubiquitous objects in contemporary research, both in the natural and social sciences. An important area of research regarding the applicability and modeling of graph- theoretical-oriented approaches to complex systems, is the probabilistic inference of such networks. There exist different methods and algorithms designed for this purpose, most of them are inspired in statistical mechanics and rely on information theoretical grounds. An important shortcoming for most of these methods, when it comes to disentangle the actual structure of complex networks, is that they fail to distinguish between direct and indirect interactions. Here, we suggest a method to discover and assess for such indirect interactions within the framework of information theory.
Information-theoretical measures (in particular, Mutual Information) are applied for the probabilistic inference of complex networks. Data Processing Inequality is used to find and assess for direct and indirect interactions impact in complex networks.
We outline the mathematical basis of information-theoretical assessment of complex network structure and discuss some examples of application in the fields of biological systems and social networks.
Information theory provides to the field of complex networks analysis with effective means for structural assessment with a computational burden low enough to be useful in both, Biological and Social network analysis.
Complex networks, no doubt constitute one of the cornerstones of contemporary research in many branches of science (Barabási 2012; Newman 2003). From systems biology to the study of the role of friendship in the spread of diseases, connections between individuals at different organizational levels are outlined using complex graphs.
Alongside with the statistical analysis of large complex networks, the need for robust methodologies to infer such networks from empirical data has risen. Most of these methods rest on the domain of probabilistic inference and computational learning (Bickel and Doksum 2007; Hernández-Lemus and Rangel-Escareño 2011) and as such, they are subject to expectation errors and asymptotic constraints. Apart from the problem of inferring large networks from noisy data sources (Bansal et al. 2007; de Jong 2002; Hernández-Lemus et al. 2009), complex network researchers in general, are confronted with some subtler structural challenges in network reconstruction. One of such challenges lies in the capacity to assess direct from indirect interactions (Chua et al. 2008; Tresch et al. 2007).
The assessment of direct and indirect interactions may play an important role in understanding network navigability, community structure and information flow, specially when considering the range of influence of given nodes within a network, the strength of the interactions and how these interactions shape the whole correlation structure, as well as the topology and dynamics of the system. For instance, in gene regulatory networks, when it comes to the functional role of subnetworks forming either motifs or pathways, there is an important distinction between genes locally involved in the regulation of a small set of highly specific targets and some other genes that are involved in the transcriptional control of a large number of targets, often by means of a chain of indirect interactions. The first set of genes is responsible for the fine tuning processes involved in environment-specific genomic control, while the second set (known as master regulators) is related with long range, large scale control of genome expression used by the cell mechanisms of growth and proliferation (Baca-López et al. 2012).
In the case of social networks, there is also a growing interest in the role that indirect interactions may play in information and influence flow among nodes (Fowler and Christakis 20072010). In some instances (such as the social epidemiology of obesity) (Fowler and Christakis 2007) it has been shown that indirect connections (i.e. second degree links) within a social network may, under some conditions, exert a greater influence than direct interactions that however shape the global structure of the network. Such structural determination may be one of the keystones to discern between diverse features of influence in social networks such as homophily, social contagion and covariation (Shalizi and Thomas 2011). We may envisage other instances in the realm of complex networks where the distinction between direct and indirect links may be of some relevance, such as ecological or economic-financial networks, etc (Beltrán et al. 2012; Callaway and Howard 2007; Tsatskis 2012).
Methods for direct and indirect interactions assessment found in the reviewed literature were designed ad hoc for too dense or too small networks (Nawrath et al. 2010; Yan et al. 2007), and most of them require additional information in order to estimate or tune parameters to differentiate the two kinds of interactions (Chua et al. 2008; Yan et al. 2007). Noteworthy is to say, a great deal of importance was expressed in direct and indirect interactions assessment, regardless the particular field of research (Systems Biology, Economics or Ecology). Most efforts invested in distinguishing between these two kinds of interactions among nodes were either done manually (for instance, see (Beltrán et al. 2012; Callaway and Howard 2007)) or rely on extremely specific issues of the underlying networks (Baldazzi et al. 2010; Nawrath et al. 2010), hence the relevance of the method we introduce in this paper.
We suggest a general method to discover and assess for direct and indirect interactions within the framework of information theory. The method we submit in this paper allows to reconstruct the basic structure of complex networks. Since our method rests on the comparison of Mutual Information (MI) among nodes in a triangle, it is not affected by directionality between links, directionality is detected instead, once the basic structure of the network is already in place. In what follows we will discuss some methods based in probabilistic inference and information theory by means of which researchers may be able to infer and assess complex network structure from the probability distributions of some empirical quantitative features.
Information theory (IT) offers a powerful theoretical foundation that is well-fit to contribute to the development of computational methodologies intended to deal with network inference problems as applied to real data in several branches of complex systems theory (Hernández-Lemus and Rangel-Escareño 2011). IT also provides an analogy with statistical mechanics (SM), that can be useful for inferring network interactions (links) from between-node correlation measures, thus enabling to use (although in a quite non-trivial manner) the huge arsenal of tools of this science. There are, however a number of open questions in the application of IT to the probabilistic complex network inference. The applied algorithms may be able to return intelligible models relying on scarce a priori information while dealing with a potentially large number of variables. IT methods may also detect non-linear dependencies in highly noisy non-independent probability distributions. The best benchmarking options for such kind of complex network inference, for us, seems to be the use of sequential search algorithms (instead of stochastic search, typically involving the assignment of structures for large constrained datasets, since these procedures have a high computational complexity, even NP-hard" -exponentially large search-space-) and performance measures based on IT, since this makes feature selection fast end efficient, and it also provides an easy way to communicate results.
Information theoretical measures have been applied intensively to infer interactions in complex networks, in particular in the field of computational biology (Bansal et al. 2007 de Jong 2002 Fleuret 2004 Hernández-Lemus et al. 2009 Margolin et al. 2006 Peng et al. 2005 van Someren et al. 2002) but also in social network studies (Crowley-Riddey 2009 Dong 2011 Mislove 2009 Mislove et al. 2010 Zhao et al. 2011). A group of correlation measures including mutual information, Markov random fields and Kullback-Liebler divergences, amongst others are considered appropriate to perform probabilistic network inference (Hernández-Lemus and Rangel-Escareño 2011). However, since conditional probabilities obey the so-called tower property, a number of false positives links may appear as a consequence of indirect correlations (Hernández-Lemus and Rangel-Escareño 2011).
For instance, if node (or agent) A has a high value of conditional correlation (say, mutual information) with node B, and B is also highly correlated with node C, most common algorithms would predict (with a marginal probability p ind ) the presence of a (possibly non-existent) link between processes A and C. In order to correct for the presence of indirect links we may implement some methods from IT, such as bounds in the information-theoretical probability measures and the use of the Data Processing Inequality (DPI) (Sehgal et al. 2007). DPI can provide a bound to the extent on which signal processing may optimize probabilistic inference. We will discuss these and other ideas in the framework of network inference and structure assessment. We will also discuss some of their implications, and potential applications in the contemporary complex systems scenario.
Some of the essential notions of IT that will be used in this work are: (information-theoretical) entropy, mutual information and other related measures. To do so, let X and Y denote two discrete random variables having the following features:
Finite alphabet and respectively
Joint probability mass distribution p(X, Y)
Marginal probability mass distributions p(X) and p(Y)
Conditional probability mass distributions p(X|Y) and p(Y|X)
Equality holds iff X and Y are statistically independent.
A comprehensive catalogue of algorithms to calculate diverse information theoretical measures (including mutual information) has been developed for [R], the statistical scientific computing environment (INFOTHEO 2012). We will analyze the special role that MI has in the field of complex networks inference from quantitative feature data. MI has been applied successfully as a measure to infer 2-way interactions in complex networks (quite specially in the field of Gene Regulatory Networks or GRNs) (Andrecut and Kauffman 2006a; Andrecut and Kauffman 2006b; Madni and Andrecut 2007; Margolin et al. 2006). As we have seen, MI quantifies the degree of statistical dependency between two random variables (say α and β). One can see that MI(α, β) = 0iff α and β are statistically independent.
Hence, if we measure some quantitative-feature of interest ϑ (say expression level of genes in GRNs), by studying its profile (and more specifically the mutual correlation profile for a set of nodes) we may find interactions conforming a network. A pair of agents characterized by feature distributions ϑ i and ϑ j for which MI(ϑ i , ϑ j ) ≠ 0 are said to interact with each other. Since MI is reparametrization invariant, one usually calculates the normalized mutual information. In this case MI(ϑ i , ϑ j ) ∈ [0, 1], ∀i, j.
With these definitions in mind, let us consider two random variables, X and Y, whose mutual information is MI(X, Y). Now consider a third random variable, Z, that is a (probabilistic) function of Y only. It can be shown that PZ|X Y = PZ|Y, which in turn implies that PX|Y Z = PX|Y, as follows from Bayes’ theorem.
An information-theoretical theorem called the Data Processing Inequality (DPI) states that Z cannot have more information about X than Y has about X; that is MI(X;Z) ≤ MI(X;Y). We can see that MI(X;Z) = H(X) − H(X|Z) ≤ H(X) − H(X|Y, Z) = H(X) − H(X|Y) = MI(X;Y). Inequality follows because conditioning on an extra variable (in this case Y as well as Z) can only decrease entropy (in a similar way to what occurs in statistical physics when adding constraints to a thermal system), A formal definition of such a theorem would be:
Three random variables X, Y and Z are said to form a Markov chain (in that order) denoted X → Y → Z, if the conditional distribution of Z depends only on Y and is independent of X. i.e. if we know Y, knowing X doesn’t add anything new to what we already know about Z than if we know only Y.
By the Markov property, since X and Z are independent, given Y, MI(X;Z|Y) = 0, then, since MI(X;Y,Z) ≥ 0 we have: MI(X;Z) ≤ MI(X;Y) c.q.d. □
There are similar approaches to the one just presented, for instance the ones in reference (Liang and Wang 2008) and in reference (Zhang et al. 2012). Both approaches are based in conditional mutual information (i.e. the degree of information a variable X and a variables Y share, given a third variable (or group of variables) Z). These algorithms try to account for indirect links by means of conditioning the associated mutual information distributions. In reference (Zhang et al. 2012), the authors further improve the performance of the algorithm by using an adaptive computation framework. In addition to these good performance general methods (references (Liang and Wang 2008; Zhang et al. 2012) were developed for gene regulatory networks although with minimal adjustments can ba applied to any other probabilistically inferred networks), there are also more specific approaches based on somehow Ad Hoc considerations. We can mention, for instance the MARINa algorithm (Lefebvre et al. 2010; Lefebvre et al. 2007; Mani et al. 2008) developed specifically for the assessment and reconstruction of gene regulatory networks based on statistical enrichment of certain signatures (Subramanian et al. 1554), an approach close in philosophy of that of conditioning variables that, however requires for additional information (i.e. the signatures themselves) to be useful, hence is more restricted to its scope and applications as are approaches relying on additional phenotypic information (Wu et al. 2009; Yu et al. 2006).
In order to introduce the importance of telling direct network interactions from indirect ones, we performed a topological analysis on the Gene Network of the fruit fly, Drosophila melanogaster (D.m.). The fruit fly Gene Network is a paradigmatic system for genetic studies and one of the best annotated organisms in genomics databases. It also presents a high genomic similarity to that of mammals -humans included- (about 61% of disease-associated genes in humans have a D.m. counterpart) and there is open access to its high-throughput inferred biological network (Costello et al. 2009). By discussing some features of the network structure of this highly studied species we introduce the problem of finding direct and indirect interactions in complex networks inferred from experimental data. Once this problem has been outlined, we proceed to illustrate how the methods of information theory may be appropriate to distinguish between direct and indirect interactions in order to sketch (at least partially), the network structure on a gene regulatory network inferred from experimental data obtained from 1191 whole genome gene expression experiments in breast tissue from breast cancer patients/controls and on a social network inferred from researchers at Mexico’s National Institute of Genomic Medicine coauthorship collaborations data, retrieved from the PubMed database.
As explained before, the methods of information theory used here correspond to the implementation of MI calculations and DPI to infer and prune respectively such networks. There is a number of different methods for computing this quantities in the literature (Hernández-Lemus and Rangel-Escareño 2011) and most of them are quite functional and almost equivalent in performance. Here we used the C++ implementation of the aracne algorithm (in particular we resort to aracne 1.0 even if there is a new version 2.0 with an improved algorithmic complexity performance, because version 2.0 uses a bootstrapping method that we have found to be still a little bit unstable) (Margolin et al. 2006) for Biological Networks and Python scripts (some customized and others from the NetworkX library) for the Social Networks. The aracne 1.0 algorithm is useful for our purposes since it is based on crystal clear MI calculations (Hernández-Lemus and Rangel-Escareño 2011), it is possible to implement DPI thresholds and its algorithmic complexity and performance are quite good (we have benchmarked aracne 1.0 against other information-theoretical methodologies such as Information Based-Similarity (ibs) and linear correlation predictors in the past (Hernández-Lemus et al. 2009) with very acceptable results). Cytoscape and Python’s library NetworkX were used to depict and analyze the networks (Assenov et al. 2008).
Microarray pre-processing of the data was performed by using the affy library in BioConductor running under [R] on a 128 Gb RAM 8-Power5+ dual core-processor, symmetric multiprocessing (SMP) unit by IBM. All statistical tests were performed on a Dell Precision Series 16 Gb RAM QuadCore Workstation by using limma package in [R]/BioConductor. Information theoretical measure calculations for biological systems were performed by the aracne v 1.0 program in the IBM SMP machine. Python scripts were used instead for Social network calculations. Graphical depiction and network analyses were performed on a MacBook Pro 8 Gb i7.
The Drosophila melanogaster GRN used to highlight the presence of hierarchical structure was not further used here to demonstrate IT methods of network assessment. The reason for this is that it was inferred (Costello et al. 2009) by using Pearson correlation metric, which is a linear measure, thus unable to capture the whole statistical dependency spectrum. Let us recall that for two statistically independent random variables Pearson correlation coefficient is 0. However, the converse is not always true, because Pearson correlation coefficient detects only linear dependencies between two variables. Null Pearson correlation coefficients only implies statistical independence for the special case of jointly normal distributions. Since this is not the general case in gene expression distributions, values of linear correlations are not enough to determine statistical dependency (Hernández-Lemus et al. 2009).
In order to introduce the need for complex network assessment of indirect interactions, let us consider the case of the gene regulatory network of the fruit fly (Drosophila melanogaster, D.m.), which is one of the best curated biological networks that has been constructed ever. By means of computational integration in a probabilistic model of high throughput data sets from DNA, RNA and protein assays, combined with data mining for individual genes to phenotypes treats, such network has been assembled with the goal of providing a ...meaningful, functional gene network and to draw new and unforeseen connections...(Costello et al. 2009). D.m. is of course a model organism for too many instances in genetics and genomics. One particular feature that is long known for the genetic structure of D.m., is the fact that gene regulation is largely determined by the action of a relatively small set of molecules that are able to exert transcriptional control in the highly active stages of development and proliferation (Baker 2001; Brennecke et al. 2007; Harrison et al. 2011; Wells 2009). For instance, aspects of regeneration are often regulated by complex mechanisms of activation for several growth regulatory pathways in damaged tissue. It has been proved that every pathway involved is being regulated by the p53 molecule in D.m., (dP53). dP53 is thus a critical master regulator in the GRN of D.m., in particular with regards to cell growth, proliferation and differentiation. Interestingly enough, the human homolog of this very molecule is known to play a quite important role in most human cancers. dP53 is by no means the only master regulator in drosophila; an extremely important gene that acts as a universal master regulator in D.m. is the molecule called eyeless (ey). This gene was first study in relation to eye development (hence the name, larvae with knocked-up eye does not develop eyes). However, the role of ey is not restricted to eye development. Abnormal expression of the gene is able to convert other tissues into eye-cells, including legs, wings and antennae. ey is also involved in controlling the processes that coordinate differentiation of several cell types in a very precise way in order to develop the fly eye. The ey-homolog Pax6 and homologs of other eye determination genes from D.m., are also required for tissue development in vertebrates. It has been shown that ey becomes a master regulator of development deep until later stages by means of regulation by signaling through the Notch and EGFR signaling pathways (two outstanding important generalistic signaling pathways) (Baker 2001).
Other master regulator genes are known to play special roles in D.m. Such is the case of the zinc-finger protein Zelda (ZLD) that plays a key role in transcriptional activation and as a master regulator of genome activation in the earliest stages of D.m. development (Harrison et al. 2011). The ZLD transcription factor bounds to thousands of sites across the genome at all developmental stages of D.m., with relatively small changes in binding between stages. The number and range of ZLD targets - around 2,000 genes have ZLD bound to their promoters and/or enhancers - demonstrate that it plays a major role in maternal-to-zygotic transition activation. Hence Zelda should be an important hub in the transcriptional regulation network for D.m. However, probably the most important group of master regulators in the fruit fly GRN is formed by the so-called piwi family of transcription factors. The piwi class of genes was identified long ago as encoding regulatory proteins that are responsible for the maintenance of incomplete differentiation in stem cells and also in preserving the stability of cell division rates in germ line cells, a quite important set of processes related again to all stages of development, growth and differentiation (Brennecke et al. 2007). There are plenty of other examples (Drosophila melanogaster is known to have some 50 or more master regulators in its GRN (Costello et al. 2009) but we think that these may be sufficient to establish the point that within the GRN for D.m. the presence of such highly connected nodes (hubs) should be evident.
In order to test the value of IT methods for community inference in the context of social networks, we describe a Scientific Collaboration Network (SCN from now on) based on the coauthorship history of the researchers of the National Institute of Genomic Medicine of Mexico (referred as INMEGEN). INMEGEN is one of the National Institutes of Health in Mexico and it was created in 2005, being so the second youngest institute of all 13.
We paid special attention to general values such as number of nodes N, clustering coefficient C, centrality degree, network centralization, and characteristic pathways 〈l〉. Network 1 has N = 847, a very high clustering coefficient 〈C〉 = 0.925 (see Figure 6C) and a network centralization of 0.221. The characteristic path length 〈l〉 is 3.09 (Figure 6E), and the average number of neighbors is 24.8. Network 2 has N = 535, a 〈C〉 = 0.411 (Figure 7C), and a network centralization of 0.199. The characteristic path length 〈l〉 is 3.36 (Figure 7E), and the average number of neighbors is 4.3. Network 3 displayed the following values, N = 535, 〈C〉 = 0.341 (Figure 8C), and a network centralization of 0.197. The characteristic path length 〈l〉 is 3.43 (Figure 8E), and the average number of neighbors is 4.1.
Due to the shortness of the characteristic path length and its clustering coefficient, the structure of Network 1 is close to that of a small-world network, which is not at all strange to collaboration networks and to SCN, as has been noted before (Yousefi-Nooraie et al. 2008). When compared to Network 2, its clustering coefficient decreases abruptly compared to its value in Network 1, and the characteristic path length increased slightly. Given the difference between the clustering coefficients of these two networks, it lead us to suggest that external collaboration plays a great deal in defining the small-world structure of INMEGEN’s collaboration network, and most of all, it gives it a closure that makes it easily navigable. By closure we mean that INMEGEN, being such a young institution, it may still depend, to a degree, on external collaborators. Such dependency might be fueled by the fact that INMEGEN doesn’t provide clinical services which means that it doesn’t have the possibility to systematically recruit research subjects from where to get biological samples, depending so, on other institutions for this purpose. This idea still remains to be tested and is part of our future work.
The differences between Network 2 and Network 3, are minimal. Network 3 was inferred by IT Methods for network assessment, even though, its clustering coefficient remained almost unaffected -decreased from C = 0.411 in Network 2, to C = 0.341 in Network 3- and the characteristic path length was also basically the same. This is an important result because it means that most of the clustering was due to external collaboration as it was noted already, and for Networks 2 and 3, triangle formation depends on having at least two nodes from INMEGEN.
IT Methods applied to Network 2, eliminated what was left from the ’unbalanced’ triangles and the weakest link between two nodes in a triangle was deleted, generating Network 3. This method made possible to identify more clearly the existence of four communities that we could recognize as INMEGEN’s main research groups, with a leader (laboratory or researcher leader) easily detectable. It is important to mention that there were other, rather small groups, that have not been articulated to the main network and with a small weight value in their collaborations links. These are hypothesized to be groups in process of consolidation.
In order to analyze the structure of the four groups, we created a subnetwork for each one, based on researchers that have collaborated in at least two publications. The subnetworks made also possible to recognize different collaborative strategies among them. Group 1 and 2, with an N = 79 and N = 139 respectively, were very similar, both displayed a high value for network centralization, over 0.750 and low clustering coefficient, having Group 2 the highest 〈C〉 = 0.356. For these groups, there was a leader with a very high k, and one or two very close collaborators. The difference between first and second order collaborators was large, with the highest weight (corresponding approximately to k = 9) between the leader and its first order collaborator and a lesser weight (implying k = 2) with the lowest, with an average connectivity difference of 7. The structure of these two networks could be interpreted as having a tendency towards a star-shaped structure.
In group 3, with N = 106, network centralization was not as high as in groups 1 and 2, although it remained important (Network centralization= 0.681). Noteworthy was the clustering coefficient over 〈C〉 = 0.500. According to these numbers, group 3 behaves more like a group and the weight of collaborations is similar among its members.
Group 4 had the highest network centralization 0.938 as well as the highest clustering coefficient 〈C〉 = 0.746. The network of group 4 depends on two members of INMEGEN on which all collaborations are fixed. The situation for group 4 is that from all, this is the extreme case of group behavior since there is a minimal connectivity degree difference of 1 between first and second order collaborators, that is, for most nodes, the weight of collaboration among each other is the same (weight 2) and only few have a weight of 1, and none of weight 3 or higher. It is important to mention that this group is different from the other three because is the newest and the smallest (N = 64).
We created these networks -i.e. Networks 1, 2, and 3- in such a way because we wanted to have a complete image of the collaborative network of INMEGEN, and to assess INMEGEN’s internal community structure. Overall, there are some emerging groups and senior researchers are still quite scarce. This may be the consequence of genomics being a new field (as compared, for instance to other biomedical disciplines), and INMEGEN is still in its infancy. If this is true, we will be able to see it in our future work when we compared INMEGEN’s SCN with the networks of other National Institutes of Health, some of them with more than 50 years of history. We would also like to model the collaboration strategies followed by the different groups using agent based modeling.
When reconstructing the basic structure of a network, to being able to assess direct and indirect interactions among nodes can be very useful and informative. One important aspect when analyzing complex networks is to being able to distinguish and assess direct from indirect interactions (even more, with regards to local interaction levels that may shape the large scale structure and the functional features of such networks). In this paper we have shown how an application of simple theorems of information theory (Hernández-Lemus and Rangel-Escareño 2011) makes possible for researchers to grasp the nature of the links in a clear-cut way. We discuss this in the context of both biological and social networks. However, we believe that this same general method arguments may apply to any network inferred by means of mutual information measures, and to some extent to other networks inferred by other quantitative interaction measures. What is more, such a general method may serve to the general purpose of unveiling similarities and differences between networks that are as different as it is a transcriptional network from a collaborative one. It is important to recall that DPI-pruned networks may be considered along with non-DPI pruned (and even with relatives degrees of pruning as given by different values of DPI tol ) in order to assess for structural features of the network. This is specially relevant if one is to consider the de-Novo functional discovery of the role of specific individuals or the reassurance of already envisioned hypothesis along the same lines.
The material presented here is intended to show some useful features of the application of IT methods for network inference and assessment. The biological and social networks chosen here were of different nature: different in size and, more important, in structure. This may be seen by the fact that DPI-pruning of the biological network (that was much larger than the social one) resulted in a tree-like structure (with no triangles and hence with a null clustering coefficient), whereas in the social network, DPI-pruning uncover different structural properties due to the presence of undecidable cases where three edges were equally weighted. However, in both cases one important outcome of direct-vs-indirect interaction analysis resulted, i.e. the presence (or better the visibility of the presence) of key players: in one case transcription factor genes acting as master regulators and in the other researchers identified as group leaders. Both studies revealed some surprising elements -like the presence of novel master regulators, or some in-consolidation researchers that are really emerging as emerging group leaders- given the structure of the direct interaction networks.
The study of the relationship between topological structure and functional organization in complex networks is, of course, still in a very early stage of development. The problems and challenges that arise include not only the determination of direct and indirect interactions and the role that some privileged nodes may play in network structure and navigability. A thorough study may also considered the issues of community structure and local connectedness. In the case of networks probabilistically inferred from experimental data one may also take into account the role that inference errors, noise and asymptotics may play. Further on, since complex networks are often dynamically adaptive systems, driven both by their inner structure and their environmental constraints, the role that dynamic evolution, fluctuations and adaptability may play will most surely also be fundamental to understand their behavior. We can only envisage what lies in the future: too many challenges should still be tackled before reaching a complete understanding of the behavior complex networks. However, we believe that information theoretical concepts and tools may play a fundamental role when facing such complexities.
EHL is a theoretical physicist trained in chemical physics, probability theory and non-equilibrium statistical mechanics; his main area of research is the interface of systems biology, statistical physics and complex system’s theory. JSG is an anthropolgist/ethnologist trained in theoretical biology and philosophy; his current research is in social networks; science, technology and society- studies; and genome-bioethics.
The authors gratefully acknowledge support by grants: 179431/2012 (CONACYT) and PIUTE10-92 (ICyT-DF) [Contract 281-2010], as well as federal funding from the National Institute of Genomic Medicine (México).
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.